HEWLETT-PACKARDJOURNAL
MAY1970
© Copr. 1949-1998 Hewlett-Packard Co.
Design and Application
of Silicon IMPATT Diodes
DC power is converted directly into microwave energy by an
IMPATT diode. In X-band, these devices generate higher CW
power than any other present-day, solid-state microwave source,
clearing the way for new cost-sensitive microwave designs.
By A. M. Cowley
MICROWAVE POWER GENERATION has been the last major
holdout against the solid-state revolution. The smaller
size, longer life, and lower power consumption of solidstate circuits could make a host of startling developments
possible but, just as vacuum tube designers reached a
frequency ceiling with grid-controlled vacuum tubes,
solid-state designers are faced with an apparent limit to
the capabilities of transistors. Yet, just as vacuum-tube
designers succeeded in generating microwave power us
ing new concepts in electron beam control, solid-state
designers are now generating microwaves by taking ad
vantage of certain ways charge carriers behave in semi
conductors.
One technique for solid-state generation of microwave
energy was suggested a dozen years ago by W. T. Read of
the Bell Telephone Laboratories.1 Read proposed that
the finite delay between an applied voltage and the cur
rent generated by avalanche breakdown in a speciallydesigned PN junction device, plus the transit delay as the
charge carriers travel through the diode,2 could amount
to a total delay of 180° at microwave frequencies. Such
a mechanism would thus contribute power to an external
circuit.
Practical realization of a "Read" diode, however, was
not achieved until seven years later3 despite the fact that
Read presented a complete and basically correct analysis
of the device. The problem was largely thermal: how to
remove the heat generated in the device. Since microwave
applications require very small devices for practical im
pedance levels, and since operation in avalanche break
down at the required current densities means extremely
high power densities — as much as 10s watts per cubic
centimeter — , a method had to be found to extract large
quantities of heat efficiently to prevent destructively high
temperatures.
A successful thermal design was achieved by inverting
a mesa-type PN junction diode and bonding the top of
the mesa with heat and pressure directly to a gold-plated
copper heat sink, as shown in Fig. 1 (solder wouldn't do
here because of its thermal resistance). When all the
variables in the thermal-compression bonding process
are at the optimum, thermal resistances low enough to
permit continuous microwave oscillations are obtained.
Cover: Scarcely an eighth
of an inch long, IMPATT di
odes generate more than
one-half watt of microwave
power in X band (8-12.4
GHz) when imbedded in a
resonant circuit supplied
with dc power. The article
beginning on this page
explains the why's and
wherefore's of these mighty
Lilliputians.
In this Issue:
Design and Application of Silicon
IMPATT Diodes, by A. Michael Cowl
e
y
Measuring Capacitance Automati
cally, by Hitoshi Noguchi, ladeo
Shimizu and Koichi Maeda . .
O HEWL
PRINTED IN U.S.A.
© Copr. 1949-1998 Hewlett-Packard Co.
page 2
page 14
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 1. IMP ATT diode formed on silicon substrate and
thermal-compression bonded to metal heat sink. Dotted
lines show shape, mechanically more stable but harder
to handle, obtained by etching from both sides of wafer.
Dimensions apply to X-band diodes.
Thermal-compression bonding, though, takes more
time than conventional techniques (e.g., soldering) and
it requires extreme care to produce an adequate and re
producible bond. In a search for a less expensive and
more consistent means of constructing an adequate
thermal path, members of the development team at the
HP Associates Division of Hewlett-Packard developed
a means of electroplating a heat sink on the junction side
of a wafer, and then etching mesa diodes directly on the
heat sink,4 as shown in Fig. 2. Because the integral heat
sink diffuses the diode heat over a relatively wide area,
the individual diodes can then be soldered into a package
or circuit by conventional techniques.
Thermal resistances typically between 14 and 17°C/W
are obtained in Hewlett-Packard's X-band diodes made
this way. Power output for the two types of HP diodes
are shown in Fig. 3. A typical diode can generate 700
mW at 10 GHz with 6.5% efficiency at a junction tem
perature of about 200°C. One watt output can be ob
tained at a junction temperature of 250 to 260°C. Since
burnout does not occur until a junction temperature of
350°C is reached, there is a large margin for safety (until
extensive reliability studies are completed, however, con
servative operation at junction temperatures below
200°C is recommended).
The new HP diodes, in common with most avalanche
diodes intended for microwave oscillators, use a simple
PN junction in a three-layer P"NN* structure rather than
' This method was proposed independently by Bernd Hoefflinger (see reference [5]).
the special four-layer N+PIP+ structure proposed by
Read. The three-layer structure has all the microwave
oscillation properties of the four-layer structure but is
easier to make. Both of these devices are called IMPATT
diodes, an acronym derived from IMPact ionization and
Avalanche Transit Time.
Among the important benefits added by the integral
heat-sink construction technique are a reduction in fab
rication costs and a significant improvement in uniformity
from diode to diode. Also, since the PN junction is at
the wide base of the mesa, rather than at the narrow top,
the slope of the mesa sides reduces the electric field at
the junction periphery, lowering the chances of pre
mature edge breakdown. Furthermore, the diodes can be
passivated by conventional dielectric deposition tech
niques while still in wafer form.
HP IMPATT diodes made with integral heat sinks are
currently priced at $150 each in sample quantities, as
compared to the $200-$300 price range of earlier
IMPATT's. There is every reason to believe that the
price will go down substantially as both demand and
production rates for these diodes grow.
Hence, the era of solid-state microwave systems is
upon us. IMPATT diodes, presently the highest power
solid-state source of CW microwave power in X-band,
are already being designed into airborne Doppler naviga
tion radars, point-to-point microwave communications
links, and airborne phased array radars. Besides making
Fig. 2. Integral heat sink serves as wafer substrate dur
ing manufacture of new HP IMPATT diodes. Chips can
be soldered into circuit or package without seriously
degrading thermal properties.
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 3. Typical performance of
HP IMPATT diodes in X-band.
Two types are currently offered.
possible some rather complex microwave systems, the
IMPATT diodes will also contribute to low-cost, com
pact systems like radar transponders, navigation beacons,
collision avoidance systems, disposable ECM jammers,
fuses, automobile radar braking systems, and intrusion
alarms.
Theory
To provide insight into IMPATT diode performance
in preference to exact quantitative design theory, several
simplifying assumptions are made in the description of
diode operation presented in this section. The basic theory
was well described by Read in his original paper1 and
more recently Sze" and deLoach7 have reviewed the
theory in the light of experimental results.
A diagram representing a three-layer P+NN+ diode is
shown in Fig. 4. The diode is represented as biased into
reverse breakdown, which gives rise to a positively
charged depletion layer that has a triangular field distri
bution, the field maximum being at the PN junction.
Breakdown occurs when the electric field is so high that
an electron or hole entering the depletion region has a
significant probability of ionizing a hole-electron pair. If,
on the average, each carrier in the depletion region pro
duces just one new hole-electron pair, a sustained ava
lanche can exist, in much the same way an electrical dis
charge occurs in a gas. The diode can thus support any
dc current with essentially a constant voltage Vbd, the
breakdown voltage.
The probability of ionization is a strongly increasing
function of the electric field. For this reason, charge
generation in a diode biased into breakdown tends to be
localized near the P+N junction, i.e., in the region of
highest electric field. In an abrupt silicon P+N junction,
this region, known as the avalanche zone (Wa in the
diagram), is about 25% of the total depletion layer
width. The remainder of the depletion layer is the drift
zone, shown as Wd.
When the diode is operating, charge is generated
throughout the avalanche zone. Under the influence of
the high electric field, the charge drifts to the N+ end of
the depletion region at constant velocity. It is charac
teristic of this mechanism that the charge velocity is
independent of electric field levels above 2 x 104V/cm;
under typical IMPATT operating conditions, the field is
higher than this over most of the depletion region. The
resulting charge velocity, known as the saturated velocity,
vsat, is about 107cm/s for electrons in silicon.
Since diode operation involves the time delay in the
avalanche process, we need to consider how charge builds
up in response to an applied voltage. From analysis of
charge build-up in a thin avalanche zone in a diode biased
with a dc voltage just equal to the breakdown voltage Vbd,
and with a sinusoidal microwave voltage superimposed,
Read arrived at an approximation that shows a sharp
pulse occurring just as the total diode voltage drops
below the breakdown level, as shown in Fig. 5. The
charge pulse occurs in the avalanche region and as the
charge moves across the drift zone, it generates a rec
tangular current waveform in the external circuit. The
© Copr. 1949-1998 Hewlett-Packard Co.
Avalanche Generation
Takes Place Here
Where Field is Highest
quency, as explained above. In practice, this is done by
careful control of the resistivity and consequently the
breakdown voltage of the N-type layer. The N-type re
gion, however, should not be much thicker than required
to support the depletion layer at breakdown, since "un
depleted" material contributes to series resistance.
When epitaxial wafer requirements are met, the re
maining parameter to be controlled is junction area, since
it determines the operating or depletion capacitance, Cd,
of the diode. Depletion capacitance is related to junction
area, A, and depletion layer thickness, Wsc, by:
A
pF
Ws
where A is in cm2 and Wsc is in cm.
If Wsc were 4 X 10-4cm (4.0 p.m) and A were 2
X 10-4cm2, Cd would be 0.5 pF. In principle, maximum
power output is increased as junction area A is made
larger, but diode reactance is reduced. Convenience of
circuit design usually dictates a capacitance that gives
a reactance around 20 to 30 ohms at the operating
frequency.
€,,=
Neutral (undepleted)
Zone
Contributes to Rs
Fig. 4. Representation of diode with abrupt P*NN* ¡unc
tion reverse-biased in avalanche breakdown. Avalanche
is generated in W* region where electric field is highest.
Charge carriers move at constant velocity through drift
region Wd. W!C is total space charge region.
duration of the external current waveform is determined
by the time T required by the charge to move across the
drift zone, and is equal to Wd/vsat. If T equals one-half
the ac period, the fundamental Fourier component of the
current waveform will be 180° out of phase with the
diode voltage. The diode thus delivers ac power to the
external circuit or, looking at it from another viewpoint,
the diode has negative resistance at this frequency.
External Circuit Current Due
to Charge (Electrons) Drifting Across Drift Zone
Diode Design
Fabrication of IMPATT diodes at Hewlett-Packard
is described step-by-step in the box on page 8. The wafer
is designed so that transit time across the drift region is
roughly one-half a period of the desired oscillation fre
Fig. 5. Charge builds up in avalanche zone just as micro
wave voltage falls below breakdown level. Movement of
charge across drift zone causes rectangular current
waveform in external circuit.
© Copr. 1949-1998 Hewlett-Packard Co.
assuming penect spreading in mtinite copper
heat sink and ~5 m thickness of silicon
Fig. 6. Thermal resistance of
HP IMPATT diode vs. ¡unction
area. Smaller diodes are closer
to ideal than larger ones.
Thermal resistance of Hewlett-Packard IMPATT
diodes was measured using the dc incremental avalanche
resistance technique, a variation of a method originally
proposed by Haitz8. Some typical results are shown in
Fig. 6. The solid line represents the ideal — the best one
can do with a silicon mesa bonded to a copper heat sink.
It was calculated from the assumption that the thermal
resistance consists of perfect thermal spreading resistance
working into a semi-infinite copper heat sink, plus the
contribution to thermal resistance of the silicon itself.
The diodes with small junction areas come very close
to the ideal. The larger area diodes do not come as close
because the solder interface with the package assumes
greater importance. For the Hewlett-Packard X-band
diodes, which have junction areas around 2 XlO~4 cm2,
the average measured values of thermal resistance are
about 15°C/W, as compared to a calculated value of
12°C/W. More detailed calculations that take into ac
count the solder interface are within a few percent of the
measured values.
P a c k a g e P a c k a g e
Inductance J ^ Capacitance
Silicon vs. GaAs
Use of gallium arsenide (GaAs) for IMPATTs has
some potential advantages, among them increased effi
ciency (23% theoretical maximum vs. 15% for Si4), and
reduced noise. These advantages are offset somewhat by
the significantly greater cost of GaAs, and the relatively
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 7. Coaxial oscillator circuit designed tor IMPATT
diode operation in X-band. Adjusting value of shunt
capacitor tunes circuit.
Poui(W);
f0 = 12.5 GHz
ldc = 130 mA
Rs = 0.8 «
\frm,(Hz);
BW = 200 Hz
fm = 10 kHz
Fig. 8. Power output and noise
vs. load resistance for two
types of IMPATT diodes (fm is
spacing of measured 200-Hz
band from oscillator center fre
quency).
primitive state of GaAs material and device technology
compared to that of Si. It is likely that GaAs IMPATT
diodes will find their place in those high-performance
applications where greater cost can be tolerated; silicon
devices are destined to be inexpensive, and hence will
appeal to designers of low-cost, high-volume products.
Making Cool-Running
IMPATT Diodes
P-N junction
A. Construction starts with highly-doped N silicon wafer of
low resistivity (=0.01 il cm). On this, first an N layer (p =
0.5 - 1.0 n cm) and then a P* layer (p = 0.003 il cm) are
grown epitaxially. Double epitaxial construction achieves
abrupt junction and closely controlled layer widths.
B. Wafer is metallized on junction (P*N) side by thin-film
techniques. Copper or gold is then electroplated on metallic
film to thickness of 0.003 to 0.004 inch.
C. Wafer's other side (N*) is thinned mechanically and
chemically to reduce wafer thickness to 0.001 to 0.002
inch. This side is then metallized.
D. Standard photoresist techniques are used to etch metal
into circular pads and to etch silicon mesas (mesa diameter:
0.006 inch).
E. Wafer is cut apart. Relatively large metal substrate left
with each mesa is soldered to copper stud of HewlettPackard style 41 microwave package. Gold strap or wire is
thermal-compression bonded to top of mesa and to top edge
of package. Package interior is then filled with inert gas
and, finally, Kovar or steel cap is welded to top, forming
hermetic seal.
© Copr. 1949-1998 Hewlett-Packard Co.
Metalization
Plating
Fig. 8 b.
Oscillator Circuit
corresponds to constant ac voltage across the diode.
The behavior shown in Fig. 8 is qualitatively explained
by a simple modification of Read's basic large-signal
theory9.
RF power output as a function of bias current is shown
with R,, as a parameter in Fig. 9. Here, it is seen how
small values of RL lead to saturation at relatively low
values of I,k. and power output. The output power level at
which saturation occurs becomes higher as RL is in
creased up to a particular level. Note in Fig. 9 that in
each case, the diode voltage at saturation is about the
same: ¥2 V,,,,. It should be emphasized that this is not a
thermal effect — it is purely electrical. (The diode used
for the data in Fig. 9 is typical of some of the better
diodes obtained during our development program: for
RL = 2.2n, 0.8W was obtained at a junction temperature
of 195°C with 10W input; for RL == 2.6Q, LOW was
obtained at T¿ = 240°C and 12.7W input).
The oscillator circuit used to obtain results reported
here is shown in Fig. 7. It is simply a length of 7 mm
coaxial line with the diode mounted at the end in a clamp
for good heat sinking. Inductive loading is provided by
the diode package inductance and by the length (1) of
line between the shorted end and the shunt radial capac
itor CL. This circuit also transforms the 5 On line imped
ance to an impedance of 1 to 3 ohms as seen by the diode.
The circuit is tuned by inserting appropriate dielectric
shims between the two parts of capacitor CL. To load an
X-band diode properly, the value of CL lies typically be
tween 1 and 3 pE
The equivalent circuit shown in Fig. 7b is trans
formed into an equivalent series-load circuit in Fig. 7c.
The values of RL and XL in Fig. 7c were calculated and
have also been measured with the Hewlett-Packard
Model 8542A Automatic Network Analyzer. It was
found that RL and XL could be calculated within 10%
of the measured values.
Power output and noise as a function of RL are plotted
for two diodes in Fig. 8. These show behavior typical of
all diodes tested: as RL is reduced from a relative high
value, say 3 ohms, the power increases slowly until "satu
ration" a self-limiting effect, takes place. Saturation oc
curs when the peak value of the applied ac waveform, VD
sin wt, reaches a value equal to one-half the diode break
down voltage, VM. The power then drops linearly as a
function of a further drop in RL. This linear dependence
Noise
When voltage saturation occurs or is imminent, the RF
spectrum becomes noisy, and under certain conditions
bias current oscillations can occur. This situation is
shown in the frequency spectra illustrated in Fig. 10.
Here, a low value of RL was chosen so that saturation
would occur at a relatively low power, 0.4 watt. The
photos show that as the bias current is increased beyond
the saturation level, the spectrum degrades rapidly.
9
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 9. Typical power output as
a function of bias current with
load resistance as a parameter.
1
Power limits when VD = — - vbd.
Measuring IMPATT Diode
Thermal Resistance
vs Vrt, at Constant
Temperature (Slope = 1/Rsc)
Thermal resistance measurements made during this study
relied on the fact that the breakdown voltage Vbd of silicon
PN junctions is almost an exact linear function of tempera
ture. For example, the abrupt P-N ¡unctions in HP IMPATT
diodes have a breakdown voltage vs. temperature charac
teristic aVbd/ST sr 1.17Vbd mV/°C. If Vbd were 70 volts at
25°C, Vbd would thus increase 82 mV for each degree of
temperature rise in the junction.
With reference to the diagram, the junction voltage Vdo
can be considered as having three components: Vbd(T0),
-^i AT, and IdcR.c. VM(To) is the room-temperature break3T
AT is the increment in Vbd caused by
down voltage,
oT
junction heating, and ldcRSc is the voltage drop in the junc
tion space charge resistance, Ric.
Rie, a purely electronic effect, can be measured by
straightforward pulse techniques. With R!C known, deter
mination of the thermal resistance « is simply a matter of
accurately measuring the junction voltage and current.
Derivation of 9 from these quantities is as follows:
A7 = T - T, = e
.) +
V* =
g
- (T - T,) +
=
37
V d c / *
This method is fast, reliable, and accurate. It has a
further advantage in that 9 is determined under actual op
erating conditions.
10
© Copr. 1949-1998 Hewlett-Packard Co.
X-317A-30
400 600 1000
Fig. 9b.
In Fig. lOd, a resonant circuit had been connected
from the bias port to ground. With the oscillator operat
ing above the saturated power level the bias circuit oscil
lated, introducing AM and FM modulation on the RF
output, as shown by the sidebands in the frequency
spectrum.
The noisy behavior accompanying power saturation
appears to be characteristic of IMP ATT diode opera
tion. For quiet operation, then, an IMPATT diode should
be operated well below the saturation level, i.e., with RL
somewhat larger than the value that obtains maximum
power. Values of FM and AM noise measured in diodes
operated this way are plotted in Fig. 1 1 . In view of the
circuit Q's employed, these are respectable noise per
formances. At the time of writing there has been no
concerted effort to characterize FM noise in circuits with
higher Q. Isolated measurements, however, in cavities
with Q's on the order of 100 to 200 at 350 mW power
levels, have shown FM less than 7 Hz (rms) in a 100-Hz
bandwidth in X-band. This figure is competitive with
reflex klystrons and better than that of BWO's found in
typical laboratory signal generators.
Modulation
Amplitude modulation of IMPATT diodes can be
achieved by directly modulating the diode's bias current.
This method also produces some frequency modula
tion simultaneously. Frequency modulation can also be
achieved with YIG or varactor circuits, or by modulating
the frequency of an external injection-phase-locking sig
nal. This technique is especially suitable when an injec
tion-locking scheme is already included for frequency
stabilization. A typical application is in microwave
telecommunication systems where an injection-locked
IMPATT oscillator serves as the transmitter.
Conclusions
The IMPATT diode has been shown to be a practical
source of CW microwave power in X-band. The funda
mentals of IMPATT circuit design are well understood in
terms of simple large-signal equivalent circuits. New
manufacturing techniques produce diodes reliably at rea
sonable cost, pointing the way to large-scale use of these
devices in microwave systems design. Future plans at the
HP Associates Division of Hewlett-Packard include de
velopment of diodes for operation at frequencies both
above and below X-band, from C band through Ku band,
and further applications work focused on those areas
which appear especially suitable for IMPATT diodes.
11
© Copr. 1949-1998 Hewlett-Packard Co.
ldc = 55 m A; P0 = 176 mW
f0 = 8.2 GHz
RL = 1.0 n
b ldc = 90 mA; P0 = 395 mW
c ldc = 95 mA; P0 = 400 mW
I = 90 mA;
(RESONANT BIAS CIRCUIT)
Fig. with Spectra of IMPATT diode oscillator operated at 8.2 GHz with 1.0 n load. Vertical
scale is 10 dB/cm. Horizontal scale is 1 MHz/cm in a, b, and c; 30 MHz in d.
Acknowledgments
understanding of IMPATT diode noise behavior. Val
uable insight into the large-signal diode operation was
contributed by Zvonko Fazarinc and Chu Yen of the
Hewlett-Packard Laboratories. Finally, during the devel
opment stage diodes could never have been made without
the skills and perseverance of R. C. Patterson, M. R.
Gavette, G. Horn, M. Bryan, and E. Petherbridge. S
Overall guidance and direction through most of the
IMPATT diode development program was provided by
Robert A. Zettler. The simple coaxial cavity oscillator
design was suggested by Robert D. Hall. Steven H. Ham
ilton participated in the later stages of circuit develop
ment and also made significant contributions to an
12
© Copr. 1949-1998 Hewlett-Packard Co.
References
[1 ]. W. T. Read. Jr., 'A Proposed High-Frequency, NegativeResistance Diode; Bell System Technical Journal, Vol. 37,
No. 2. March 1958.
[2]. W. Shockley, 'Negative Resistance Arising from Transit
Time in Semiconductor Diodes; Bell System Technical Jour
nal, Vol. 33, No. 4, July 1954.
[3]. R. L. Johnston, B. C. deLoach, Jr., and B. G. Cohen.
'A Silicon Diode Microwave Oscillator; Bell System Tech
nical Journal, Vol. 44, No. 2. February 1965.
[4]. R. A. Zettler and A. M. Cowley. 'Batch Fabrication of
Integral Heat Sink IMPATT Diodes; Electronics Letters,
Vol. 5, No. 26, December 27, 1969.
[5]. B. Hoefflinger, 'Recent Developments on Avalanche
Diode Oscillators; Microwave Journal, Vol. 12. No. 3.
March 1969.
[6]. S. M. Sze, 'Physics of Semiconductor Devices; Chapter
5. Wiley-Interscience, New York. 1969.
[7]. B. C. deLoach. in 'Microwave Semiconductor Devices
and their Circuit Applications; H. A. Watson. Ed.. McGrawHill, New York, 1969.
[8]. R. H. Haitz, H. L. Stover. N. J. Tolar, 'A Method for
Heat Flow Resistance Measurements in Avalanche Diodes;
IEEE Transactions on Electron Devices, Vol. ED-16, No. 5,
May 1969.
[9]. A. M. Cowley, Z. Fazarinc. R. D. Hall, S. H. Hamilton.
C. Yen. R. A. Zettler. The Singly-Tuned IMPATT Oscilla
tor; presented at the IEEE Solid-State Circuits Conference,
Philadelphia, February, 1970.
A. M. Cowley
Mike Cowley earned BSEE and
MSEE degrees at Notre Dame
(1959 and 1961) anda PhD
in electrical engineering at
Stanford in 1965. Along the way
he taught electrical engineering
at the University of California
at Santa Barbara and worked as
a consultant in physical
electronics. He joined HP in
1965, where he has been
concerned with Schottkybarrier diode technology and
with microwave oscillators
using avalanche effects in
silicon and the Gunn effect in gallium arsenide. Mike is
presently manager of the Applied Research Section of the
HP Associates R and D Laboratory. He is a member
of the IEEE, Tau Beta Pi, Sigma Xi, and the American
Physical Society.
Staunch members of the Sierra Club, Mike and his wife
enjoy backpacking in the high Sierras with their two
sons, aged 1 and 21/2 years, usually with the help of a
not-too-stubborn burro.
13
© Copr. 1949-1998 Hewlett-Packard Co.
Measuring Capacitance Automatically
Completely automatic multifrequency
bridge measures capacitance from
0.001 pF to 1.2 fjiF, conductance
and dissipation factor.
By Hitoshi Noguchi, Takeo Shimizu
and Koichi Maeda
VALUES OF CAPACITANCE USED IN CIRCUIT DESIGN range
from less than 1 pF to 1 F. To measure capacitance and
loss factor over this wide range, designers most often
use the frequencies 1 20 Hz, 1 kHz or 1 MHz to drive a
capacitance bridge.
Large capacitors (greater than l,uF) and electrolytics
are usually measured at the lowest frequency. Capacitors
less than 1 ^F down to several hundred pF are mea
sured at 1 kHz, and capacitors less than about 1000 pF
are commonly measured at 1 MHz. Most capacitance
bridges use a single frequency — either 1 kHz or 1 MHz
— for capacitance and loss measurements.
Efficiency in capacitor measurement can be improved
with automatic methods of measurement. Measurements
at several frequencies rather than a single frequency also
allow complete evaluation of frequency sensitive devices.
The new HP Model 4270A Automatic Capacitance
Bridge, Fig. 1, automatically measures capacitance and
loss at four frequencies between 1 kHz and 1 MHz. It
measures capacitance from 0.001 pF to 1.2 /.iF in six
automatically selectable ranges. The loss component is
measured as either parallel conductance or dissipation
factor. The loss (G) can be measured down to 0.1
nanomhos. The instrument has an internal, continuously
variable dc bias supply to 200 volts to measure param
eters of such devices as varactor diodes.
Balancing the Bridge
A bridge measurement, obviously, is valid only when
the bridge is balanced. Various types of bridges are
balanced in different ways. The best bridge design bal
ances quickly with relatively simple operation. Ideally,
the null detector should control bridge operation during
balancing.
14
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. measure grounded new HP Model 4270 A Automatic Capacitance Bridge can measure grounded
cables L- input capacitance of amplifiers with the LOW terminal grounded. In LGROUND operation, the instrument still retains the guarded feature.
AC bridges often use phase detectors as null detectors.
It is a characteristic of ac bridges that the bridge output
waveform inverts as the bridge output passes through
the null.1 Thus, the phase detector can provide informa
tion about the direction of balance. In the Model 4270A,
this information through logic circuits, directly controls
the balancing operation.
In the Model 4270A, the output of the phase detector
is integrated. Detector sensitivity is a function of the ingration time interval. When the bridge is far off null, the
integrating time is short (1 millisecond) for rapid ap
proach to the null. But detector sensitivity is lowered. As
the null is approached, integrating time is automatically
increased to about 20 milliseconds, and detector sensitiv
ity is increased. Thus the bridge is balanced in the fastest
possible time.
The traditional, manually-operated bridge uses a very
sensitive null detector. The detector, however, shows only
the magnitude of the unbalanced voltage or current, but
does not indicate the direction necessary to achieve bal
ance. Satisfactory control of the bridge depends largely
upon the operator's experience.
Basic Operation
The Model 4270A Bridge is balanced by adjusting
attenuators in the C and D,G arms (Fig. 2). Digital
signals operate transistor switches in each attenuator,
Fig. 3. When the proper value of attenuation is reached,
the currents ix, k-, and iD or i0 equal zero and the bridge
is balanced. (The capacitance and loss displays indirectly
15
© Copr. 1949-1998 Hewlett-Packard Co.
FUNCTION • CAPACITANCE • D.G
S W I T C H D I S P L A Y D I S P L A Y
C ,
D , G
SWITCHING • REVERSIBLE • REVERSIBLE
CONTROL B COUNTER & • COUNTER &
D R I V E R D R I V E R
CURRENT
DETECTOR
PHASE
â € ¢
TIMING
PULSE
GENERATOR
C Balance
D.G Balance
E r r o r
Signals
DIGITAL
LOGIC
CIRCUIT
Fig. 2. capacitance logic circuits control range and balance functions of the capacitance bridge.
indicate attenuator resistance values.) During balancing,
the phase detector output, through the logic circuitry,
operates the reversible counter and driver in the direc
tion necessary to achieve balance.
Balance Equations
Null detector current i(1 is the sum of current ix through
the unknown, current Lr through the C measurement arm,
and current iD or iG through the D, G measurement arm.
From the block diagram, Fig. 2,
ix = (Gx + ¡<»Cx)ex = (Gx + joCx)Axe0.
Since the gain Ax of operational amplifier Cx has been
selected to be 1, then
= (Gx + j<oCx)e0
Fig. 3. Transistor switches in the attenuator (51, etc.)
ground resistances to keep noise low.
(1)
16
© Copr. 1949-1998 Hewlett-Packard Co.
play digit through the reversible counter is opened. Whether
the counter runs forward or reverse depends upon the po
larity of the comparator output. Digital signals from the
counter change the attenuators and operate the display
once during each integrating interval. When the integrator
output polarity changes (which means that the signal is off
null in the other direction) the gate to the next digit is
opened, and the counter reverses. Attenuators are operated,
and each digit actuated in sequence by repeating this
process.
If the integrator output drops below the reference level
while sequencing through the C display, or when the last C
digit is reached and integrator output polarity changes, the
"parameter change" circuit switches to D,G. Each D,G digit
is then actuated in sequence by repeating the same process.
If the integrator output drops below the reference level while
sequencing through the D,G display, the null circuit stops
the measurement. Otherwise the measurement stops after
the last digit has been actuated.
Logic Circuits
The logic circuits are the control circuits that balance the
bridge. They switch the standards and digitally select values
of attenuation so that balance is reached and maintained.
When a measurement is started, the gates that operate the
C$ and Gs relays (through the range selector) are opened in
the "Auto Range" mode. The "parameter change" circuit
switches to C.
The signal from the phase detector is integrated as de
scribed in the text. From the integrators, the signal goes to
comparators that determine its polarity and compare its
level with a reference voltage. Polarity is an indication of
which direction the signal is off null; level is an indication
of how close the signal is to the null.
Two gate controls, one for C and one for D,G are actuated
by the comparator output. The gate controlling the Cs and
G, relays is closed, and the gate to operate the first C dis
Digital Signals
For Cs Attenuators
And C Display
Digital Signals
For G8 Attenuators
And D,G Display
Control Signals
For Cs, G5 Relay
PARAMETER
CHANGE CIRCUIT
'Bridge Balanced' Signal Stops Measurement
17
© Copr. 1949-1998 Hewlett-Packard Co.
Similarity
Ã- c = jiaCsec = j<aCaAce0
lo = Gse0 = GsAUi0e0
(2)
At bridge balance, id = 0. Then i<, = ix + ic + ÃG = 0.
T h u s
C x -
â € ” A C C S
GX = — AD¡GGS
To measure dissipation factor D, the slide switch on the
front panel is moved to D. Substituting source voltage
e'o (see Fig. 2) for source voltage e,,,
I'D = GseD = GsAD¡Be'0 = G,AüiGAi,er (3)
Since id = ix + ic + ÃD = 0, from equations (1), (2)
and (3),
_ GX
— GsAutoADAr'
Gs ¿ ^
o)C.\
[n the Model 4270A,
Fig. 4. Feedback admittance in the current detector is
adjusted automatically to maintain constant bridge sensi
tivity over the operating range.
- is selected to be ^— and AD =
— 27T
resistance is thus kept constant, regardless of the attenua
tor setting.
Stray capacitances can cause differences in the time
constant between the attenuator resistors. This appears as
error in measuring dissipation factor or conductance
especially at 1MHz. In the Model 4270A, the attenua
tors are shielded, and the highest value resistors have
adjustments so that time constant differences can be kept
below 1 00 picoseconds.
10 '
Thus,
D x —
A o, G
(4)
At balance then, Cx = — AcCs, Gx = — AD,0G8 and
Dx = — AD, Q.
As mentioned previously, the front panel readouts are an
interpretation of attenuator settings, which set the gains
AC and ADj0 in the above equations.
Three-terminal measurements reduce errors introduced
by stray capacitance. In a conventional ac bridge, a
transformer isolates the source voltage so that threeterminal measurements are possible. But transformers
have limited frequency range. Therefore, another scheme
was used in the Model 4270A to cover the 1 kHz to
1 MHz frequency range with only one bridge network.
In the Model 4270A, each measurement arm consists
of a resistance attenuator and an operational amplifier.
All of the amplifiers have identical frequency and phase
characteristics. All use low-impedance elements to reduce
effects of stray capacitance at high frequencies.
All components in the resistance attenuator are like
wise low-impedance elements. The resistance attenuators
are a form of digital-to-analog converter. Digital signals
actuate the switching transistor, adding or removing re
sistances, and producing an analog output voltage.
Switching transistors are in series with the standard
resistors in the attenuator. It is possible to compensate
for any deviation in each resistor by adjusting the base
current of its switching transistor. Switching is so ar
ranged that unused resistances are grounded. The output
Current Detector
When using the guarded three-terminal method of
measurement, terminal-to-ground capacitance will de
crease the sensitivity of a voltage sensitive null detector.
With a phase sensitive null detector, this capacitance
causes phase shift which also reduces the detection capa
bility. A solution to this problem, used in the Model
4270A, is the current sensitive detector and preamplifier
with low input impedance.
The detector has a feedback circuit, Fig. 4, consisting
of an adjustable capacitor and resistor. Values of the
feedback capacitance and resistance are adjusted to (1)
maintain constant detector sensitivity, (2) maintain a con
stant phase relationship between the bridge driving volt
age e(J and the detector voltage ed, and (3) maintain a low
input impedance to the detector preamplifier. This as
sures stable bridge operation and reduces problems asso
ciated with phase detection.
Because the input impedance to the detector preampli
fier is low. the voltage at that point is low. Thus the effect
of stray capacitance is negligible. Also, the low input
18
© Copr. 1949-1998 Hewlett-Packard Co.
SPECIFICATIONS
HP Model 4270A
AUTOMATIC CAPACITANCE BRIDGE
MEASURING CIRCUIT:
FLOAT: Guarded terminals of unknown are floated from ground.
L-GROUND: One side of unknown terminals is grounded, guard is re
tained.
STANDARD CAPACITANCE:
PARAMETERS MEASURED: Capacitance, equivalent parallel conduct
ance and dissipation factor.
MEASURING FREQUENCY: 1 kHz, 10 kHz, 100 kHz and 1 MHz, ±1%.
AVAILABLE FULL SCALE RANGES:
CAPACITANCE
CONDUCTANCE
10 kHz 100 kHz
1 MHz
18.000pF
999. 9n
180.00pF 18.000pF
9.999ii
1800. OpF 1 8 0 . 0 0 p F 1 8 . 0 0 0 p F
99.99^
18.000nF 1800.0pF 180.00pF
999. 9ft
ISO.OOnF 18.000nF 1200.0pF
9.999m
DISSIPATION
FACTOR
0.9999
REPEATABILITY: ±2 digits at NORMAL TEST VOLTAGE, ±8 digits at
LOW TEST VOLTAGE.
RANGE MODE:
AUTO: Range selection and balance performed automatically.
HOLD: Range is held on fixed position determined by previous AUTO
and can be selected manually by stepping RANGE STEP. Balance
begins with fourth digit from right.
TRACK: Range held on fixed position, balance begins with last digit.
BALANCING TIME: Typically 0.5 s.
MEASURING RATE: Measurement cycle equals balance time plus dis
play time. Balance time typically 0.5 s. Display times selected by
MEAS RATE are 70 ms, 2 s, 5 s, and MANUAL.
RESET: Bridge is reset and new balance operation started automatically
(at MEASURING RATE), or manually by RESET pushbutton, or re
motely by contact closure. If the unknown changes value during the
balance operation, or other changes shift the unknown outside of the
balance range, the bridge resets to rebalance after 2 seconds.
TEST VOLTAGE ACROSS UNKNOWN:
NORMAL: 500 mV rms constant, at capacitance units displayed in pF
or nF, 100 mV rms constant at /iF.
LOW: 100 mV rms constant at pF or nF. 20 mV rms constant at ¡if.
Other test voltages are available.
ACCURACY:
CAPACITANCE, CONDUCTANCE AND DISSIPATION FACTOR:
DC BIAS: INTERNAL or EXTERNAL to ±200 V, in HOLD and TRACK
mode.
INTERNAL BIAS at FLOAT measurement:
VOLTAGE: 0 to 20 V de; O to 200 V dc; continuously variable on
front panel, monitored on rear panel.
DIAL ACCURACY: ±5% of full scale.
SOURCE RESISTANCE: 100 kfl.
POLARITY: LOW unknown terminal (-), HIGH unknown terminal
( + ) in FLOAT position of MEAS CKT control.
REMOTE: Programmable by resistor with 250 Ã.VV rate at 20 V range,
25 fi/V rate at 200 V range.
REMOTE ACCURACY: ±2% of full scale.
INTERNAL BIAS at L-GROUND: An additional connection using a
blocking capacitor and a coaxial cable is necessary for INTERNAL
source.
DATA OUTPUTS: (If unknown in FLOAT)
CODE: Positive 4 line BCD (1-2-4-8), 13 columns consisting of decimal
location, overrange, units, 5 digits of capacitance and 4 digits of
conductance or dissipation factor.
'1' LEVEL: +5 V, open circuit, source Impedance 7.5 kSi at all data
except at 10« digit of C data and OVER-RANGE. +4V, open circuit,
source impedance 7.5 k2 at 10' digit of C data and OVER-RANGE.
'0' LEVEL: Ground.
PRINT COMMAND: DC coupled, print level + 5 V, print holdoff level
0V.
INPUTS:
TRIGGER HOLD OFF LEVEL: Level must be between -10 V and
-15 V.
REMOTE PROGRAMMING: Eight front-panel functions can be remotely
controlled by external contact closure to ground with impedance
less than 400 Q. Programmable functions are RESET, FREQUENCY,
RANGE MODE, TEST VOLTAGE, LOSS MEASUREMENT, RANGE
STEP, DC BIAS RANGE, BIAS VERNIER.
OPERATING TEMPERATURE: 0°C to 50°C.
POWER REQUIREMENTS: 115 or 230 V ac ±10%, 50 to 60 Hz, approx.
110 W.
WEIGHT: 34 Ib (15.5 kg).
PRICE: $4825.00.
NOTE: Cs = internal standard capacitor. Cx = capacitance measured.
Dx = dissipation factor measured.
FREQUENCY: ±1%.
MANUFACTURING DIVISION: YOKOGAWA-HEWLETT-PACKARD LTD.
9-1, Takakura-cho
Hachioju-shi
Tokyo, Japan
© Copr. 1949-1998 Hewlett-Packard Co.
impedance nearly eliminates noise induced by electro
static coupling, mainly power line noise. Capacitance
introduced by connecting cables may be balanced out
with an available option.
Applications
For development of new components, circuits or sys
tems, the Model 4270A can be operated as an indepen
dent lab-type instrument. It can be used to measure
discrete capacitor values, cable capacitance and loss,
input and output capacity, strays, junction capacitance
and other non-linear capacitances. Other areas using the
instrument alone include quality assurance, production or
incoming inspection.
The HP Model 4270A has been designed for use with
other instruments, computers or in systems. Data out
puts (measured value, multiplier and units) are available
in BCD form. Signals for remote programming are sup
plied to the bridge through a rear-panel connector. All
major functions can be controlled individually. Thus,
the Model 4270A finds application in quality control,
reliability analysis, and statistical analysis of products.
Users are able to make comparison checks, environmen
tal classification, sorting and inspection. In-process testing
of integrated circuits is another important application.
There are also applications where capacitance itself
is not the main parameter being measured, but ca
pacitance is monitored to derive other information. For
example, in process control, dielectric or capacitance
variation may be monitored. Changes in capacity or loss
indicate changes in other physical parameters.
Acknowledgments
The authors wish especially to acknowledge the con
tributions of the following people: Giichi Yokoyama, who
proposed the original idea of 4270A; Hiroatsu Kohno
who did the mechanical design; and Kazunori Shibata
who handled the industrial design phase of the project.
Other contributors include Masakatsu Shida, Arimichi
Nakagiri and Saburo Sugihara. S
Reference
[1]. K. Yoshimoto, 'A New Universal Impedance Bridge;
Hewlett-Packard Journal, Sept. 1966.
Hitoshi Noguchi
Hitoshi Noguchi obtained a
BSEE degree from Akita
University in 1961. He then
joined Yokogawa Electric
Works as an R and D engineer,
working on the development
of signal generators. Hitoshi
joined Yokogawa HewlettPackard in 1964, working on
the design of the Model 4260A
Universal Bridge, and then on
the design of the Model 4270A
as the project leader.
Takeo Shimizu
Takeo Shimizu received his
BSEE degree from Fukui
University in 1961, and joined
Yokogawa Electric Works as
an R and D engineer, working
on the development of a dc
potentiometer, dc bridge, and
mechanical strain measure
ment. Takeo joined YHP as a
senior field engineer in 1964. In
1966 he transferred to the
R and D section and developed
logic section circuitry for the
HP Model 4270A.
Koichi Maeda
Koichi Maeda received his
BSEE from Waseda University,
Tokyo, in 1 963. After a year of
working at Yokogawa Electric
Works as an R & D engineer, he
joined YHP in 1964. At YHP
he worked on oscillator devel
opment, specifically on the
design of the Model 4204A
Digital Oscillator. He designed
the bridge section circuitry
for 4270A.
HEWLETT-PACKARDJOURNAL_©MAY1970volume21•Number9
TECHNICAL CALIFORNIA FROM THE LABORATORIES OF THE HEWLETT-PACKARD COMPANY PUBLISHED AT 1501 PAGE MILL ROAD. PALO ALTO. CALIFORNIA 94304
Editor: Danielson A Jordan Editorial Board: R. P. Dolan, H. L. Roberts, L. D. Shergalis Art Director: Arvid A Danielson Assistant: Maridel Jordan
© Copr. 1949-1998 Hewlett-Packard Co.